A Growing Model-Based OCSVM for Abnormal Student Activity Detection from Daily Campus Consumption

With the rapid development of information technology, Smart Campus Card System (SCCS) has become an important part of digital campus construction. Although the characteristics of students' abnormal activities can be reflected in smart campus card records, studies now focus more on analyzing the relationship between normal student activities and smart campus card data. Therefore, we analyzed some features of students’ amounts of consumption on campus and their statistical characteristics, and established a dataset based on smart campus card records for the purpose of detecting students’ abnormal activities. However, extant anomaly detection methods are prone to the two issues listed below. First, the vast majority of existing unsupervised anomaly detection algorithms are trained by fitting a central piece of the training data while disregarding the anomalous data. Those approaches do not completely eliminate the impact of anomaly data. Secondly, those algorithms have poor time performance on large-scale datasets. This paper proposed a growing model-based one-class support vector machine (GMB-OCSVM) to solve the above problems. Our model outperforms other frontier models in terms of detection accuracy and execution efficiency after extensive testing. In particular, our method processes 61,000 more units of data per second than the OCSVM method in terms of execution efficiency and improves detection accuracy to 92.39%. It demonstrates that our method can effectively handle the challenges of abnormal data interference in the training process and inefficient execution, and can detect abnormal student behavior in the campus daily consumption dataset in an efficient and accurate manner, indicating that our method has some practical utility.